Demographic characteristics of respondents and influence of moderating factors
Two studies (Nissenbaum et al., 2012 and Shepherd et al., 2011) did not provide suitable data for statistical pooling. One of these KX2-391 (Nissenbaum et al., 2012) reported the “near group” as having significantly worse sleep scores for both PSQI (p = 0.046) and ESS (p = 0.03); and two subjects in the “near group” were diagnosed with insomnia compared to none in the “far group”. In the second study (Shepherd et al., 2011), participants with greater exposure to WTG noise reported significantly worse sleep scores (p = 0.0006). For the remaining six studies which provided suitable data, three (Bakker et al., 2012, Pedersen and Persson, 2004 and Pedersen and Persson, 2007) used low SPL values of < 30 dB as controls, while two (Krogh et al., 2011 and Magari et al., 2014) compared groups based on the distances of respondents' from WTGs. Meta-analysis revealed a cohesion significant increase in the odds of reporting sleep disturbances with greater exposure to noise from WTGs (OR 2.94; 95% Cl: 1.98 to 4.37; I2 = 0%; p < 0.00001; Fig. 3). Subgroup analysis by SPLs or distance did not result in a change in the direction of the results. A similar result was observed when five studies with higher respondents' rates (n = 810) were meta-analysed (OR: 2.76; 95% CI: 1.65 to 4.62; I2 = 0%; p = 0.0001). Sensitivity analyses of studies with larger sample sizes (n = 838) revealed a significant increase in the odds of sleep disturbances with higher SPLs (OR: 3.24; 95% CI: 2.03 to 5.18; I2 = 0%; p < 0.00001).